[agentHits]

Problem

The dokploy service updates stop-first, so the swarm task pulled the new 4.4 GB image only after the old container had stopped. On dev the panel was down for 2 min 12 s (old container finished 11:31:42 UTC, new one started 11:33:54 UTC) and Traefik answered Bad Gateway the whole time.

Fix

• Pull first. getAgentHitsUpdateCommand and the new getOfficialUpdateCommand start with docker pull <image> || exit 1, while the old container still serves the panel. The swarm task then finds the image locally, so the downtime is just the restart (~15–20 s). If the pull fails, the service is not touched.
• Wait for the real restart. The dialog used to reload after a fixed 8 s once /api/health answered. With the pull in front, the old server still answers at that point, so the dialog would have reloaded into the old version. It now waits until the server has gone down once and is healthy again (requests time out after 5 s, gives up after 20 min with an error). Texts say the panel stays usable during the download.
• List untagged images. Images that swarm pulls by digest have no tag, and docker image ls hides them unless -a is passed. The running image after a UI update was one of them, so the image list (feat: optionally remove old Dokploy images on web server update #34) didn't show it and the cleanup would never remove such images. Both now use docker image ls -a.

Checks

• vitest __test__/settings/ __test__/security/: 253 passed (new: pull runs before docker service update in both commands, keep-count env on the official path, -a in listing and cleanup).
• tsc --noEmit in apps/dokploy; biome clean; sh -n on the generated update and cleanup scripts.
• On dev (read-only): confirmed the stop→pull→start timeline from container timestamps and docker events, and that the running digest-only image shows only with -a.
• Not run end to end: a real update through the dialog with this build.

🤖 Generated with Claude Code